Implicit conversion from float -4.599999999999994 to int loses precision